Intel looks ready to crash AMD's server market

On Monday, Intel presented its roadmap for its Opteron-killer chips, Xeon 5100 Series and explained why it will grab back any lost market share and stay ahead--this time.

The chips are strategic to Intel (INTC) in keeping its dominance in lucrative server chip markets as data centers upgrade to computing systems using less electric power; this power issue is fast becoming the the single most limiting factor to expanding any computing facility.

AMD's Opteron has managed to win widespread support because it uses less electric power and offers great performance. And it has been party time for AMD's investors as the company has finally broken into the enterprise server market.  But Intel has marshaled considerable resources to crash the party and take back any lost market share.

At a press event in San Francisco, Tom Kilroy, Intel vice president presented a broad message that Intel is going to maintain a power and performance lead over Opteron because it can make the chips smaller by using 65nm and in the near future, 45nm process technology, while AMD is still at 90 nm; it has fast buses, shared caches, and many other critical chip and system level technologies such as virtualization; and it has long established relationships with thousands of vendors, developers, customers, etc.

AMD says it is confident in building on its lead and it has certainly done well so far. But that's because it had the low-power market all to itself. Now that Intel has its new 5100 Series Xeon microprocessors and a multi-year roadmap,  things won't be as easy for AMD.

Intel's massive manufacturing organization is a full generation and more, ahead of AMD. Intel can put an advanced chip manufacturing technology such as 65nm into production faster than anyone else, and do it in three fabs at the same time, ramping to volume production in record times.

And this is Intel's core strength, its manufacturing prowess. Yes, chip designs are important but how fast you can get them into production *and* get the highest yields per silicon wafer is what always sorts out the winners in the chip industry.

AMD has to execute near perfectly on billions of dollars in investments in expanding its chip fabs while moving to 65nm processing--a challenging task in itself.

Also, AMD is a newcomer to the server chip business while Intel has long established relationships with thousands of vendors, retailers, designers and customers. And it already has an entire ecosystem of software tools etc to support its server chips. AMD needs to build similar relationships and ecosystem, which is a long, hard slog.

Intel and AMD are battling over benchmark results and many journalists and analysts are focused on making sure they use apples to apples comparisons. Nathan Brookwood, head analyst at Insight64 was advocating a type of cook-off between Intel and AMD with both optimizing their systems to their fullest degrees. That would be interesting but probably not very useful except to settle a bar bet about which is better.

Benchmarks seem to me to be a red herring in servers and designed to win the media war rather than showcase the best systems. Data center managers typically make purchasing decisions by running their own tests because of the different designs, implementations, and application loads that servers carry.

Greg Brandeau, VP at Pixar, the computer animation studio, was at the Intel event and afterwards he said he preferred plugging a system into the wall and using a commonly found wattage meter to measure electric power consumption. He said that Pixar uses Opteron systems and he would continue using them alongside the latest Xeon servers as an easy comparison. He praised Intel's software development tools which he said helped optimize Pixar's graphics software.

Also, IT enterprises have always been conservative and risk-averse. Intel represents a safer bet that it will be able to deliver on an ambitious multi-year roadmap.

AMD could cut prices to grow market share but this is a dangerous game to play against Intel, which can pump out more chips per wafer. And since chip manufacturing costs are relatively constant, more chips per wafer means a lower cost for Intel.

I'm confident AMD will maintain a decent slice of server markets because IT customers like to have a choice and so they will buy from AMD to keep Intel on its toes. The question then becomes will AMD's slice of the pie be large enough to pay for the design and manufacturing that are needed to support the Opteron family?

_ _ _